[llvm] [GlobalISel][AArch64] AArch64O0PreLegalizerCombiner: Disable fixed-point iteration (PR #94291)

[Tobias Stadler via llvm-commits]

tobias-stadler wrote:

> How are you collecting the compile_instructions metric?

I run the test-suite on Linux under an AArch64 cross-toolchain (RelWithDebInfo clang, official ARM GNU linker/libc binaries) using `-DTEST_SUITE_USE_PERF=1` , which collects the instruction count and dumps it to a file (together with wall-time). I couldn't find a way to aggregate the instruction count, so I have slightly modified the lit module that collects the compile_time: https://github.com/tobias-stadler/llvm-test-suite/commit/8d68974d60831d4a2c2180b28a6dd0e9d6f9f192
I want to file a PR for this after some more cleanup.
I do this, because instruction count is much more stable than wall-time (<+-0.1% for a single run) and also is the metric that is measured on llvm-compile-time-tracker.


https://github.com/llvm/llvm-project/pull/94291